Physical condition embedded seismic surface wave intelligent inversion method, equipment and medium

Abstract

The invention relates to a physical condition embedded seismic surface wave intelligent inversion method and device and a medium, and the method comprises the steps: obtaining the seismic surface wave data of an active source or a passive source to be inverted, and carrying out the extraction through a frequency dispersion analysis method, and obtaining an observation frequency dispersion curve of the seismic surface wave data; constructing an initial shear wave velocity model, obtaining a theoretical frequency dispersion curve and a sensitivity matrix of the initial shear wave velocity model, and calculating physical condition embedding of the observation frequency dispersion curve based on the theoretical frequency dispersion curve and the sensitivity matrix of the theoretical frequency dispersion curve; based on an initial shear wave velocity model and physical condition embedding, an inversion model is used for iterative solution to obtain an inversion shear wave velocity model, and the device and the medium are used for implementing the method. Compared with the prior art, the method has the advantages that the problem that an initial model is easy to fall into a local minimum value due to manual selection is avoided, the inversion precision of the seismic surface waves is improved, the limitation of fixed input or output dimensions is eliminated, and the applicability of the inversion method in processing actual seismic surface wave data is remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of underground exploration and imaging, and in particular to a physical condition embedded seismic surface wave intelligent inversion method, device and medium. BACKGROUND

[0002] The shallow surface is the most complex, sensitive and fragile part of the earth medium, and is also the part that is most closely related to human beings. As one of the most critical parameters of the shallow surface, the accurate and reliable determination of shear wave velocity is of great importance to the development and utilization of the shallow surface and the rational protection of the human living environment. Traditional methods for obtaining shear wave velocity, such as borehole wave velocity testing, are invasive and can only reflect local property parameters, which cannot meet the needs of rapid regional determination of the shallow surface. Seismic surface wave analysis is a non-destructive physical exploration method that uses the geometric dispersion characteristics of surface wave propagation in vertically inhomogeneous media to obtain underground rock and soil physical parameters. Since the main influencing factor of surface wave phase velocity is shear wave velocity, the shear wave velocity information of the underground medium can be estimated by inverting the observed surface wave dispersion curve. Traditional seismic surface wave inversion methods are usually physically driven, using local optimization or global optimization algorithms to obtain a shear wave velocity solution that matches the observed surface wave dispersion curve. Among them, the local optimization algorithm starts from a pre-set initial shear wave velocity model, and updates the iteration by using the first or second derivative information of the objective function with respect to the shear wave velocity model (such as gradient descent method, Newton method, etc.), in order to find a local optimal solution within the neighborhood of the initial shear wave velocity model. Therefore, this method has high solving efficiency but is sensitive to the initial shear wave velocity model. Global optimization algorithms usually draw on the heuristic information of biology or nature to search for a global optimal solution that can fit the observed surface wave dispersion curve in high-dimensional space (such as genetic algorithm, simulated annealing algorithm, etc.), but the solving efficiency is usually low.

[0003] In addition, with the development of deep learning technology in the field of artificial intelligence, its excellent nonlinear mapping capability has shown its advantages in seismic surface wave inversion. However, the methods currently used are mostly purely data-driven, and the core idea is to train a deep neural network that can directly map surface wave dispersion curves to shear wave velocity models by constructing a large-scale observation data-model parameter pair in a specific scenario. For example, Chinese Patent Application CN115407397A provides a typical purely data-driven inversion method. First, a simulation data set is constructed, and the neural network is trained using the simulation data set. Finally, the trained inversion model is used to invert the seismic surface wave. Due to the inherent differences between the simulation data and the measured data, as well as the incompleteness of the shear wave velocity model sample construction, the applicability and generalization performance are limited in actual complex scenarios.

[0004] Therefore, it is a problem to be solved to provide a high-precision, high-generalization and efficient seismic surface wave inversion method. SUMMARY

[0005] The present application aims to overcome the defects of the prior art and provides a physical condition embedded intelligent seismic surface wave inversion method, device and medium, which adopts an iterative inversion strategy from a simple homogeneous background to a complex detail description, and introduces physical condition embedding of observed seismic surface wave dispersion curves in the inversion process to solve the problem of limited accuracy of inversion results caused by relying only on pure physical driving or pure data driving in the prior art.

[0006] The object of the present application can be achieved by the following technical solutions:

[0007] According to a first aspect of the present application, a physical condition embedded intelligent seismic surface wave inversion method is provided, which comprises:

[0008] Obtaining seismic surface wave data of a primary or passive source to be inverted, extracting the observed dispersion curve of the seismic surface wave data using a dispersion analysis method;

[0009] Constructing an initial S-wave velocity model and obtaining the theoretical dispersion curve and sensitivity matrix thereof, and calculating the physical condition embedding of the observed dispersion curve based on the theoretical dispersion curve and the sensitivity matrix thereof;

[0010] Based on the initial S-wave velocity model and the physical condition embedding, an inversion model is used to iteratively solve the inversion S-wave velocity model.

[0011] As a preferred technical solution, the initial S-wave velocity model is a homogeneous half-space model with a constant S-wave velocity value.

[0012] As a preferred technical solution, the method for calculating the physical condition embedding is:

[0013]

[0014] wherein DC emb represents the physical condition embedding of the observed dispersion curve; M represents the number of frequency points of the observed dispersion curve; DC represents the observed dispersion curve; DC homo represents the theoretical dispersion curve of the initial S-wave velocity model; K homo represents the sensitivity matrix of the initial S-wave velocity model; and T represents matrix transposition.

[0015] As a preferred technical solution, the inversion model comprises an embedding layer, a plurality of encoder layers and an output layer; the encoder layer comprises a plurality of adaptive layer normalization blocks, a multi-head self-attention mechanism block, a multi-head cross-attention mechanism block and a feedforward network, and the output layer comprises an adaptive layer normalization block and a linear layer.

[0016] As a preferred technical solution, the method for obtaining the inversion S-wave velocity model comprises:

[0017] A1, the embedding layer receives a S-wave velocity model to generate an embedding vector; and the S-wave velocity model is an initial S-wave velocity model at the first iteration, and is a S-wave velocity model at the end of the (k-1)th iteration at the kth iteration;

[0018] A2, the embedding vector is added with a position encoding vector to obtain a first intermediate vector;

[0019] A3, the first vector is normalized by using a first adaptive normalization block to generate a scaling vector and a translation vector based on a current iteration number value, and the normalized first vector is scaled and translated by using the scaling vector and the translation vector to obtain a second vector; and in the first encoder layer, the first vector is the first intermediate vector, and in the lth encoder layer, the first vector is an output of the (l-1)th encoder layer;

[0020] A4, global dependency features are extracted from the second vector by using a multi-head attention mechanism block, and the global dependency features and the first vector are fused to obtain a first fusion vector;

[0021] A5, after the first fusion vector is normalized by using a second adaptive normalization block, the normalized first fusion vector is scaled and translated by using the scaling vector and the translation vector to obtain a third vector;

[0022] A6, the third vector and a physical condition embedding are taken as inputs of the multi-head cross-attention mechanism block to obtain a fourth vector, and the fourth vector and the first fusion vector are fused to obtain a second fusion vector;

[0023] A7, after the second fusion vector is normalized by inputting the second fusion vector into a third adaptive normalization block, the normalized second fusion vector is scaled and translated by using the scaling vector and the translation vector to obtain a fifth vector;

[0024] A8, the fifth vector is processed by using the feedforward network to generate a sixth vector, and the sixth vector and the second fusion vector are fused to obtain an encoder layer output;

[0025] A9. Normalizing the encoder layer output using an adaptive layer normalization block in the output layer and then combining the normalized value with the current number of iterations to obtain a second intermediate vector;

[0026] A10. Mapping the second intermediate vector to the same dimension as the shear wave velocity model using the linear layer to generate an update, and adding the update to the shear wave velocity model to obtain an output result of the inversion model;

[0027] A11. Based on the output results, invert the next iterative shear wave velocity model;

[0028] A12, determine whether the number of iterations meets the iteration end condition, if so, end the iteration, otherwise return to A1. As a preferred technical solution, the method for inverting the next iterative shear wave velocity model is:

[0029] V k+1 =V k -[G(Net(V k ,k,DC emb ),nk)-G(Net(V k ,k,DC emb ),nk-1)], where V k+1 represents the shear wave velocity model of the k+1th iteration step; V k represents the shear wave velocity model of the kth iteration step; k represents the iteration step; DC emb represents the physical condition embedding of the observed dispersion curve; Net(·) represents the inversion network processing; n is the maximum number of iterations; G(Net(V k ,k,DC emb ),nk) represents the result of smoothing the shear wave velocity model output by the inversion model at the kth iteration step nk times; G(Net(V k ,k,DC emb ),nk-1) represents the result of nk-1 smoothing of the shear wave velocity model output by the inversion model in the kth iteration step.

[0030] As a preferred technical solution, the training method of the inversion model includes:

[0031] Construct multiple elastic medium models, use the dispersion curve forward algorithm to generate the observed dispersion curve of each elastic medium model and obtain the original shear wave velocity model of each elastic medium model;

[0032] A uniform half-space model with a constant shear wave velocity is selected to obtain a theoretical dispersion curve and its sensitivity matrix, and based on the theoretical dispersion curve and its sensitivity matrix, a physical condition embedding corresponding to the observed dispersion curve of each elastic medium model is calculated;

[0033] Smooth the original shear wave velocity model of each elastic medium model by using a Gaussian filter to obtain a shear wave velocity model; and set the filling value as the constant value during the smoothing process.

[0034] Generate input data based on the shear wave velocity model of each elastic medium model, the number of iterations of the smoothing process, and the physical condition embedding, and use the corresponding original shear wave velocity model as a label, construct a training data set based on the input data and the label, and train the inversion model by using the training data set.

[0035] As a preferred technical solution, the loss function expression of the inversion model is:

[0036]

[0037] wherein N represents the number of parameters of the shear wave velocity model; V represents the label; n represents the maximum number of iterations; and k represents the index; , which is a set of iteration numbers, and the elements are iteration numbers; DC emb , which is a set of iteration numbers, and the elements are iteration numbers; DC

[0038] According to a second aspect of the present application, a physical condition embedded intelligent seismic surface wave inversion device is provided, which comprises a memory and a processor, the memory stores a computer program, and the processor implements the method when executing the program.

[0039] According to a third aspect of the present application, a computer readable storage medium is provided, which stores a computer program, and the program is executed by a processor to implement the method.

[0040] Compared with the prior art, the present application has the following beneficial effects:

[0041] 1) The physical condition embedding adopted in the present application directly maps the observed dispersion curve to the parameter space of the shear wave velocity model, and trains the inversion network by using the observed dispersion curve parameters containing the physical condition embedding, which effectively avoids the differences in the frequency band width and numerical range of the observed dispersion curve caused by using a pure data-driven inversion model for training, breaks the limitation of fixed input or output, and significantly improves the applicability of the inversion method in processing actual seismic surface wave data and the precision of the inversion of seismic surface waves,

[0042] 2), the invention adopts the iterative inversion strategy from simple uniform background to complex details: starting from the initial constant uniform half-space model, gradually embedding the guiding under the physical condition of the observation dispersion curve, using deep neural network to introduce detailed information, and finally obtaining the S-wave velocity model close to the actual complex underground medium condition; this strategy effectively avoids the problem of easily falling into local minimum value due to artificial selection of initial model, and the number of iterations is less, thereby improving the inversion efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0043] Figure 1 The method flowchart of the present application is shown in the figure;

[0044] Figure 2 The inversion model structure diagram of the present application is shown in the figure;

[0045] Figure 3 The multi-channel seismic surface wave data of the present application is shown in the figure;

[0046] Figure 4 The surface wave dispersion spectrum and dispersion curve of the present application are shown in the figure;

[0047] Figure 5 The S-wave velocity model obtained by inversion of the present application is shown in the figure. DETAILED DESCRIPTION

[0048] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should belong to the scope of protection of the present application.

[0049] Unless otherwise defined, technical terms or scientific terms used in the present application shall have the ordinary meanings as understood by one of ordinary skill in the art to which the present application pertains. The terms "a", "an", "one", "this", and similar terms as used herein do not denote a limitation of quantity but denote the presence of at least one of the referenced items. The terms "include", "comprise", "have", and any variations thereof, are intended to cover a non-exclusive inclusion, for example, a process, method, system, product, or device that comprises a list of steps or units are not necessarily limited to those listed steps or units but can include additional steps or units not expressly listed or can include steps or units inherent to such process, method, product, or device. The terms "connect", "connected", "coupling", and similar terms as used herein are not limited to a direct or physical connection but can include an electrical connection. The term "plurality" refers to two or more. The term "and / or" describes an associated relationship between associated objects, which means that there can be three relationships, for example, "A and / or B" can mean that A exists alone, A and B exist together, and B exists alone. The character " / " generally represents an "or" relationship between the associated objects. The terms "first", "second", "third", and the like are merely used to distinguish similar objects and do not represent a specific order of the objects.

[0050] To solve the problems in the prior art, the present application provides a physical condition embedded seismic surface wave intelligent inversion method, which adopts an iterative inversion strategy from simple uniform background to complex detail description: starting from an initial constant uniform half-space model, gradually introducing detailed information using a deep neural network under the guidance of the observation dispersion curve condition, and finally obtaining a shear wave velocity model close to the actual complex underground medium condition, as shown in the flowchart. Figure 1

[0051] Specifically, it comprises:

[0052] S1, obtaining seismic surface wave data of a main source or a passive source to be inverted, using a dispersion analysis method to extract the observation dispersion curve of the seismic surface wave data.

[0053] In detail, in the present application, beam analysis is used for dispersion analysis of seismic surface wave data.

[0054] S2, constructing an initial shear wave velocity model and obtaining a theoretical dispersion curve and a sensitivity matrix of the initial shear wave velocity model, and calculating the physical condition embedding of the observation dispersion curve based on the theoretical dispersion curve and the sensitivity matrix.

[0055] S21, the selected initial shear wave velocity model is a uniform half-space model with a constant shear wave velocity value, and in the present application, the constant shear wave velocity value is 500 m / s.​

[0056] S22, based on the theoretical dispersion curve and its sensitivity matrix, calculating the physical condition embedding of the observed dispersion curve, the expression is:

[0057]

[0058] Wherein, DC emb represents the physical condition embedding of the observed dispersion curve; M represents the frequency point number of the observed dispersion curve; DC represents the observed dispersion curve; DC homo represents the theoretical dispersion curve of the initial shear wave velocity model; K homo represents the sensitivity matrix of the initial shear wave velocity model; T represents the matrix transpose.

[0059] S3, based on the initial shear wave velocity model and the physical condition embedding, iterative solving is carried out by using the inversion model to obtain the inversion shear wave velocity model.

[0060] The inversion model provided by the application comprises an embedding layer, twelve encoder layers and an output layer. The input of the embedding layer is 1-dimensional, and the output is 256-dimensional. Each encoder layer comprises a plurality of adaptive layer normalization blocks, a plurality of multi-head self-attention mechanism blocks (the number of heads is 16, which is used to capture the global dependence relationship of each parameter in the shear wave velocity model), a plurality of multi-head cross-attention mechanism blocks (the number of heads is 16, which is used to fuse the observed dispersion curve information after the physical condition embedding), and a feedforward network (the size of the hidden state is 1024, the activation function is a GELU activation function approximated by using a Tanh function, which is used to capture the complex relationship between each local feature), and the output layer comprises an adaptive layer normalization block and a linear layer (the input dimension is 256, and the output dimension is 1).

[0061] The method for obtaining the inversion shear wave velocity model by using the inversion model comprises the following steps: Figure 2 as shown in the figure, comprising:

[0062] A1, the embedding layer receives the shear wave velocity model to generate an embedding vector; and the shear wave velocity model is the initial shear wave velocity model in the first iteration, and the shear wave velocity model is the shear wave velocity model at the end of the k-1 iteration in the kth iteration.

[0063] A2, the embedding vector is added to the position encoding vector to obtain a first intermediate vector.

[0064] A3, the first vector is normalized by using the first adaptive normalization block, the scaling vector and the translation vector are generated based on the current iteration number value, and the first vector after the normalization is scaled and translated by using the scaling vector and the translation vector to obtain a second vector; and in the first encoder layer, the first vector is the first intermediate vector, and in the lth encoder layer, the first vector is the output of the l-1th encoder layer.

[0065] A4, the second vector is used to extract global dependence features by a multi-head attention mechanism block, and the global dependence features are fused with the first vector to obtain a first fused vector.

[0066] A5, the first fused vector is normalized by a second adaptive normalization block, and then scaled and translated by a scaling vector and a translation vector to obtain a third vector.

[0067] A6, the third vector and the physical condition embedding are input into a multi-head cross attention mechanism block to obtain a fourth vector, and the fourth vector is fused with the first fused vector to obtain a second fused vector.

[0068] A7, the second fused vector is input into a third adaptive normalization block for normalization, and then scaled and translated by a scaling vector and a translation vector to obtain a fifth vector.

[0069] A8, the fifth vector is processed by a feedforward network to generate a sixth vector, and the sixth vector is fused with the second fused vector to obtain an encoder layer output.

[0070] A9, the encoder layer output is normalized by an adaptive layer normalization block in the output layer, and then combined with the current iteration value to obtain a second intermediate vector.

[0071] A10, the second intermediate vector is mapped to the same dimension of the shear wave velocity model by a linear layer, an update amount is generated, and the update amount is added to the shear wave velocity model to obtain an output result of the inversion model, and the output result is a shear wave velocity model.

[0072] A11, based on the output result, the next iteration shear wave velocity model is inverted.

[0073] Taking the kth iteration of the inversion model as an example, the method for inverting the next iteration shear wave velocity model comprises:

[0074] A111, the output result Net(V k ,k,DC emb ) of the inversion model is subjected to n-k times of smoothing processing and n-k-1 times of smoothing processing, respectively, and the difference between the above two times of smoothing processing is calculated.

[0075] A112, the shear wave velocity model obtained by the kth inversion is subtracted by the above difference, and the result is the shear wave velocity model obtained by the k+1th inversion.

[0076] The expression of the above process is:

[0077] V k+1 =V k-[G(Net(V k ,k,DC emb ),nk)-G(Net(V k ,k,DC emb ),nk-1)], where V k+1 represents the shear wave velocity model of the k+1th iteration step; V k represents the shear wave velocity model of the kth iteration step; k represents the iteration step; DC emb represents the physical condition embedding of the observed dispersion curve; Net(·) represents the inversion network processing; n is the maximum number of iterations; G(Net(V k ,k,DC emb ),nk) represents the output of the inversion model of the kth iteration step after nk smoothing; G(Net(V k ,k,DC emb ),nk-1) represents the output of the inversion model at the kth iteration step after nk-1 smoothing.

[0078] A12. Determine whether the number of iterations meets the iteration end condition, that is, whether the maximum number of iterations has been reached. If so, the iteration ends; otherwise, return to A1.

[0079] In addition, the present invention also provides a training method for an inversion model, comprising:

[0080] B1. Construct 20,000 elastic medium models, use the dispersion curve forward algorithm to generate the observed dispersion curve of each elastic medium model and obtain the original shear wave velocity model of each elastic medium model.

[0081] B2. Select a uniform half-space model with a constant shear wave velocity, obtain the theoretical dispersion curve and its sensitivity matrix, and calculate the physical condition embedding of the observed dispersion curve for each elastic medium model based on the theoretical dispersion curve and its sensitivity matrix.

[0082] B3. Use a Gaussian filter to iteratively smooth the original shear wave velocity model of each elastic medium model to obtain a shear wave velocity model; during the smoothing process, set the padding value to a fixed value of 500 m / s.

[0083] B4. Generate input data based on the shear wave velocity model of each elastic medium model, 20 smoothing iterations, and physical condition embedding. Use the corresponding original shear wave velocity model as a label. Construct a training dataset based on the input data and labels, and use the training dataset to train the initial inversion network.

[0084] In detail, the loss function expression of the inversion model is:

[0085]

[0086] Where N represents the number of parameters in the shear wave velocity model; V represents the label; n represents the maximum number of iterations; k represents the index; Represents a set of iteration times, whose elements are the number of iterations; DC emb Indicates the embedding of physical conditions; G(V,nk) represents the shear wave velocity model after nk times of smoothing; Net(·) represents the inversion model processing.

[0087] In order to verify the feasibility and reliability of the present invention, the following Figure 3 The multi-channel virtual source seismic surface wave data shown in the figure is generated by beamforming analysis. Figure 3 The dispersion spectrum of Figure 4 The observed dispersion curve shown by the white dotted line in the middle is extracted. The observed dispersion curve is embedded in the physical conditions and then input into the inversion network. It is iterated according to steps A1 to A12, and the final inversion is as follows: Figure 5 The shear wave velocity model represented by the dashed line is used in the same situation to collect the following data using the borehole velocity test method: Figure 5 The solid line in the middle represents the shear wave velocity model. From the two curves, it can be seen that the test results of the two methods are highly consistent, which fully illustrates the feasibility and reliability of the method provided by the present invention.

[0088] The present invention also provides a physical condition-embedded seismic surface wave intelligent inversion device, comprising a central processing unit (CPU) capable of executing various appropriate actions and processes according to computer program instructions stored in a read-only memory (ROM) or loaded from a storage unit into a random access memory (RAM). The RAM may also store various programs and data required for device operation. The CPU, ROM, and RAM are interconnected via a bus. An input / output (I / O) interface is also connected to the bus.

[0089] Many components in a device are connected to the I / O interface, including: input units, such as a keyboard and mouse; output units, such as various types of displays and speakers; storage units, such as magnetic disks and optical disks; and communication units, such as network cards, modems, and wireless communication transceivers. The communication unit allows the device to exchange information / data with other devices via computer networks such as the Internet and / or various telecommunication networks.

[0090] The processing units perform the various methods and processes described above, such as methods S1-S3, A1-A12, and B1-B4. For example, in some embodiments, methods S1-S3, A1-A12, and B1-B4 can be implemented as a computer software program tangibly embodied in a machine readable medium, such as a storage unit. In some embodiments, portions or all of the computer program can be loaded and / or installed onto the device via the ROM and / or the communication unit. When the computer program is loaded onto the RAM and executed by the CPU, one or more of the steps of methods S1-S3, A1-A12, and B1-B4 described above can be performed. Alternatively, in other embodiments, the CPU can be configured to perform methods S1-S3, A1-A12, and B1-B4 by way of other any suitable means, such as by way of firmware.

[0091] The functionality described above in this document can be performed, at least in part, by one or more hardware logic components. For example, and without limitation, illustrative types of hardware logic components that can be used include Field-programmable Gate Arrays (FPGAs), Application-specific Integrated Circuits (ASICs), Application-specific Standard Products (ASSPs), System-on-a-chip systems (SOCs), Complex Programmable Logic Devices (CPLDs), etc.

[0092] Program code for carrying out the methods of the present application can be written in any combination of one or more programming languages. This program code can be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the program code, when executed by the processor or controller, produces a means for implementing the functions / acts specified in the flowcharts and / or block diagrams. The program code can be retrieved from a machine-readable medium or device, a storage medium, a memory medium, a tangible medium, a non-transitory medium, or any suitable combination thereof. The program code can be executed by a hardware processor or controller to produce a machine that, when executing the program code, is capable of implementing the functions / acts specified in the flowcharts and / or block diagrams.

[0093] In the context of the present application, a machine-readable medium can be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable storage medium can include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of the machine-readable storage medium will include one or more lines of a processor, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0094] The above merely provides the specific implementation of the present application, but the protection scope of the present application is not limited thereto, and any skilled person in the art can easily think of various equivalent modifications or replacements within the technical range disclosed by the present application, and these modifications or replacements should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A physical condition embedded seismic surface wave intelligent inversion method, characterized in that: The method includes: Obtaining active or passive source seismic surface wave data to be inverted, extracting it using a dispersion analysis method, and obtaining an observed dispersion curve of the seismic surface wave data; Constructing an initial shear wave velocity model, obtaining a theoretical dispersion curve and a sensitivity matrix of the initial shear wave velocity model, and calculating a physical condition embedding of an observed dispersion curve based on the theoretical dispersion curve and the sensitivity matrix; Based on the initial shear wave velocity model and the physical condition embedding, an inversion shear wave velocity model is obtained by iteratively solving the inversion model.

2. The method for intelligent seismic surface wave inversion with embedded physical conditions according to claim 1, characterized in that: The initial shear wave velocity model is a uniform half-space model with a constant shear wave velocity value.

3. The method for intelligent seismic surface wave inversion with embedded physical conditions according to claim 1, characterized in that: The method for calculating the physical condition embedding is: Among them, DC emb Indicates the physical condition embedding of the observed dispersion curve; M indicates the frequency number of the observed dispersion curve; DC indicates the observed dispersion curve; DC homo represents the theoretical dispersion curve of the initial shear wave velocity model; K homo represents the sensitivity matrix of the initial shear wave velocity model; T represents the matrix transpose.

4. The method for intelligent seismic surface wave inversion with embedded physical conditions according to claim 1, characterized in that: The inversion model includes an embedding layer, multiple encoder layers and an output layer; the encoder layer includes multiple adaptive layer normalization blocks, a multi-head self-attention mechanism block, a multi-head cross-attention mechanism block and a feedforward network, and the output layer includes an adaptive layer normalization block and a linear layer.

5. The method for intelligent seismic surface wave inversion with embedded physical conditions according to claim 4, characterized in that: The method for obtaining the inverted shear wave velocity model includes: A1. The embedding layer receives a shear wave velocity model and generates an embedding vector; the shear wave velocity model in the first iteration is the initial shear wave velocity model, and the shear wave velocity model in the kth iteration is the shear wave velocity model at the end of the k-1th iteration; A2. Add the position encoding vector to the embedding vector to obtain a first intermediate vector; A3. Normalize the first vector using a first adaptive normalization block, generate a scaling vector and a translation vector based on the current iteration count, and scale and translate the normalized first vector using the scaling vector and the translation vector to obtain a second vector; wherein, in the first encoder layer, the first vector is a first intermediate vector, and in the lth encoder layer, the first vector is an output of the l-1th encoder layer; A4. Extract global dependency features from the second vector using a multi-head attention mechanism block, and fuse the global dependency features with the first vector to obtain a first fused vector. A5. Normalize the first fused vector using the second adaptive normalization block, and then scale and translate the normalized first fused vector using the scaling vector and the translation vector to obtain a third vector. A6. Embed the third vector and the physical condition as the input of the multi-head cross attention mechanism block to obtain a fourth vector, and fuse the fourth vector with the first fused vector to obtain a second fused vector. A7. Input the second fused vector into the third adaptive normalization block for normalization processing, and then scale and translate the normalized second fused vector using the scaling vector and the translation vector to obtain a fifth vector. A8. Process the fifth vector using the feedforward network to generate a sixth vector, and fuse the sixth vector with the second fused vector to obtain an encoder layer output; A9. Normalizing the encoder layer output using an adaptive layer normalization block in the output layer and then combining the normalized value with the current number of iterations to obtain a second intermediate vector; A10. Mapping the second intermediate vector to the same dimension as the shear wave velocity model using the linear layer to generate an update, and adding the update to the shear wave velocity model to obtain an output result of the inversion model; A11. Based on the output results, invert the next iterative shear wave velocity model; A12. Determine whether the number of iterations meets the iteration end condition. If so, the iteration ends. Otherwise, return to A1.

6. The method for intelligent seismic surface wave inversion with embedded physical conditions according to claim 4, characterized in that: The method for inverting the next iterative shear wave velocity model is: V k+1 =V k -[G(Net(V k ,k,DC emb ),nk)-G(Net(V k ,k,DC emb ),nk-1)], where V k+1 represents the shear wave velocity model of the k+1th iteration step; V k represents the shear wave velocity model of the kth iteration step; k represents the iteration step; DC emb represents the physical condition embedding of the observed dispersion curve; Net(·) represents the inversion network processing; n is the maximum number of iterations; G(Net(V k ,k,DC emb ),nk) represents the result of smoothing the shear wave velocity model output by the inversion model at the kth iteration step nk times; G(Net(V k ,k,DC emb ),nk-1) represents the result of nk-1 smoothing of the shear wave velocity model output by the inversion model in the kth iteration step.

7. The method for intelligent seismic surface wave inversion with embedded physical conditions according to claim 1, characterized in that: The training method of the inversion model includes: Construct multiple elastic medium models, use the dispersion curve forward algorithm to generate the observed dispersion curve of each elastic medium model and obtain the original shear wave velocity model of each elastic medium model; A uniform half-space model with a constant shear wave velocity is selected to obtain a theoretical dispersion curve and its sensitivity matrix, and based on the theoretical dispersion curve and its sensitivity matrix, a physical condition embedding corresponding to the observed dispersion curve of each elastic medium model is calculated; Iteratively smoothing the original shear wave velocity model of each elastic medium model using a Gaussian filter to obtain a shear wave velocity model; setting the padding value to the fixed value during the smoothing process; Input data is generated based on the shear wave velocity model of each elastic medium model, the number of iterative smoothing processes, and the physical condition embedding, and the corresponding original shear wave velocity model is used as a label. A training data set is constructed based on the input data and the label, and the inversion model is trained using the training data set.

8. The method for intelligent seismic surface wave inversion with embedded physical conditions according to claim 6, characterized in that: The loss function expression of the inversion model is: Wherein, N represents the number of parameters of the shear wave velocity model; V represents the label; n represents the maximum number of iterations; k represents the index; Represents a set of iteration times, with the element being the number of iterations; DC emb Indicates the embedding of physical conditions; G(V,nk) represents the shear wave velocity model after nk times of smoothing; Net(·) represents the inversion model processing.

9. A physical condition embedded seismic surface wave intelligent inversion device, comprising a memory and a processor, wherein a computer program is stored in the memory, characterized in that: When the processor executes the program, the method according to any one of claims 1 to 8 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.

Images